Systems and methods for data grafting to enhance model robustness

Abstract

An example method includes detecting, by context analysis circuitry, occurrence of a triggering condition. The example method also includes scheduling, by context analysis circuitry and based on the occurrence of the triggering condition, retraining of a model. The example method also includes generating, by data grafting circuitry and in response to scheduling the retraining of the model, a context-relevant training data set based on a target context vector. The example method also includes retraining, by model training circuitry, the model using the context-relevant training data set to mitigate deterioration of performance of the model.

What is claimed is: 
     
       1. A method for generating a context vector to be utilized as training data for retraining a model to mitigate deterioration of performance of the model, the method comprising:
 selecting, by context vector generation circuitry, a plurality of variables defining an exogenous context for a target data point, wherein selecting the plurality of variables comprises:
 determining a set of variables available corresponding to at least one of temporal or location data associated with the target data point, 
 determining a subset of variables from the set of variables that relate to the exogenous context based on a type of the target data point, and 
 selecting the subset of variables as the plurality of variables; 
 
 identifying, by the context vector generation circuitry, values for the plurality of variables based at least on a context indicator of the target data point; 
 generating, by the context vector generation circuitry, the context vector based on the identified values for the plurality of variables; 
 storing, by the context vector generation circuitry, the context vector in a known training data index; 
 identifying, by data grafting circuitry, the context vector as relevant to a target context vector associated with a current data point being processed by the model; and 
 utilizing, by model training circuitry, the context vector in connection with retraining the model. 
 
     
     
       2. The method of  claim 1 , wherein identifying the values for the plurality of variables includes:
 querying, by communications circuitry, one or more data sources using at least the context indicator for the target data point; and 
 retrieving, by the communications circuitry, at least a portion of the values for the plurality of variables from the one or more data sources. 
 
     
     
       3. The method of  claim 1 , wherein the target data point comprises a data point reflecting current information collected in near-real-time. 
     
     
       4. The method of  claim 1 , wherein the target data point comprises a data point reflecting historical information. 
     
     
       5. The method of  claim 1 , wherein the target data point comprises a data point retrieved from a preexisting data set. 
     
     
       6. The method of  claim 5 , wherein the preexisting data set comprises the known training data index. 
     
     
       7. The method of  claim 5 , further comprising generating context vectors for one or more additional data points in the preexisting data set. 
     
     
       8. An apparatus for generating a context vector to be utilized as training data for retraining a model to mitigate deterioration of performance of the model, the apparatus comprising:
 context vector generation circuitry configured to:
 select a plurality of variables defining an exogenous context for a target data point by:
 determining a set of variables available corresponding to at least one of temporal or location data associated with the target data point, 
 determining a subset of variables from the set of variables that relate to the exogenous context based on a type of the target data point, and 
 selecting the subset of variables as the plurality of variables; 
 
 identify values for the plurality of variables based at least on a context indicator of the target data point; 
 generate the context vector based on the identified values for the plurality of variables; and 
 store the context vector in a known training data index; 
 
 data grafting circuitry configured to identify the context vector as relevant to a target context vector associated with a current data point being processed by the model; and 
 model training circuitry configured to utilize the context vector in connection with retraining the model. 
 
     
     
       9. The apparatus of  claim 8 , further comprising:
 communications circuitry configured to: 
 query one or more data sources using at least the context indicator for the target data point; and 
 retrieve at least a portion of the values for the plurality of variables from the one or more data sources. 
 
     
     
       10. The apparatus of  claim 8 , wherein the target data point comprises a data point reflecting current information collected in near-real-time. 
     
     
       11. The apparatus of  claim 8 , wherein the target data point comprises a data point reflecting historical information. 
     
     
       12. The apparatus of  claim 8 , wherein the target data point comprises a data point retrieved from a preexisting data set. 
     
     
       13. The apparatus of  claim 12 , wherein the preexisting data set comprises the known training data index. 
     
     
       14. The apparatus of  claim 12 , wherein the context vector generation circuitry is further configured to generate context vectors for one or more additional data points in the preexisting data set. 
     
     
       15. A computer program product for generating a context vector to be utilized as training data for retraining a model to mitigate deterioration of performance of the model, the computer program product comprising at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to:
 select a plurality of variables defining an exogenous context for a target data point, wherein selecting the plurality of variables comprises:
 determining a set of variables available corresponding to at least one of temporal or location data associated with the target data point, 
 determining a subset of variables from the set of variables that relate to the exogenous context based on a type of the target data point, and 
 selecting the subset of variables as the plurality of variables; 
 
 identify values for the plurality of variables based at least on a context indicator of the target data point; 
 generate the context vector based on the identified values for the plurality of variables; and 
 store the context vector in a known training data index; 
 identify the context vector as relevant to a target context vector associated with a current data point being processed by the model; and 
 utilize the context vector in connection with retraining the model. 
 
     
     
       16. The computer program product of  claim 15 , wherein the software instructions, when executed, further cause the apparatus to:
 query one or more data sources using at least the context indicator for the target data point; and 
 retrieve at least a portion of the values for the plurality of variables from the one or more data sources. 
 
     
     
       17. The computer program product of  claim 15 , wherein the target data point comprises a data point reflecting current information collected in near-real-time. 
     
     
       18. The computer program product of  claim 15 , wherein the target data point comprises a data point reflecting historical information. 
     
     
       19. The computer program product of  claim 15 , wherein the target data point comprises a data point retrieved from a preexisting data set. 
     
     
       20. The computer program product of  claim 19 , wherein the preexisting data set comprises the known training data index.